1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * NVMe over Fabrics common host code. 4 * Copyright (c) 2015-2016 HGST, a Western Digital Company. 5 */ 6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 7 #include <linux/init.h> 8 #include <linux/miscdevice.h> 9 #include <linux/module.h> 10 #include <linux/mutex.h> 11 #include <linux/parser.h> 12 #include <linux/seq_file.h> 13 #include "nvme.h" 14 #include "fabrics.h" 15 16 static LIST_HEAD(nvmf_transports); 17 static DECLARE_RWSEM(nvmf_transports_rwsem); 18 19 static LIST_HEAD(nvmf_hosts); 20 static DEFINE_MUTEX(nvmf_hosts_mutex); 21 22 static struct nvmf_host *nvmf_default_host; 23 24 static struct nvmf_host *__nvmf_host_find(const char *hostnqn) 25 { 26 struct nvmf_host *host; 27 28 list_for_each_entry(host, &nvmf_hosts, list) { 29 if (!strcmp(host->nqn, hostnqn)) 30 return host; 31 } 32 33 return NULL; 34 } 35 36 static struct nvmf_host *nvmf_host_add(const char *hostnqn) 37 { 38 struct nvmf_host *host; 39 40 mutex_lock(&nvmf_hosts_mutex); 41 host = __nvmf_host_find(hostnqn); 42 if (host) { 43 kref_get(&host->ref); 44 goto out_unlock; 45 } 46 47 host = kmalloc(sizeof(*host), GFP_KERNEL); 48 if (!host) 49 goto out_unlock; 50 51 kref_init(&host->ref); 52 strlcpy(host->nqn, hostnqn, NVMF_NQN_SIZE); 53 54 list_add_tail(&host->list, &nvmf_hosts); 55 out_unlock: 56 mutex_unlock(&nvmf_hosts_mutex); 57 return host; 58 } 59 60 static struct nvmf_host *nvmf_host_default(void) 61 { 62 struct nvmf_host *host; 63 64 host = kmalloc(sizeof(*host), GFP_KERNEL); 65 if (!host) 66 return NULL; 67 68 kref_init(&host->ref); 69 uuid_gen(&host->id); 70 snprintf(host->nqn, NVMF_NQN_SIZE, 71 "nqn.2014-08.org.nvmexpress:uuid:%pUb", &host->id); 72 73 mutex_lock(&nvmf_hosts_mutex); 74 list_add_tail(&host->list, &nvmf_hosts); 75 mutex_unlock(&nvmf_hosts_mutex); 76 77 return host; 78 } 79 80 static void nvmf_host_destroy(struct kref *ref) 81 { 82 struct nvmf_host *host = container_of(ref, struct nvmf_host, ref); 83 84 mutex_lock(&nvmf_hosts_mutex); 85 list_del(&host->list); 86 mutex_unlock(&nvmf_hosts_mutex); 87 88 kfree(host); 89 } 90 91 static void nvmf_host_put(struct nvmf_host *host) 92 { 93 if (host) 94 kref_put(&host->ref, nvmf_host_destroy); 95 } 96 97 /** 98 * nvmf_get_address() - Get address/port 99 * @ctrl: Host NVMe controller instance which we got the address 100 * @buf: OUTPUT parameter that will contain the address/port 101 * @size: buffer size 102 */ 103 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size) 104 { 105 int len = 0; 106 107 if (ctrl->opts->mask & NVMF_OPT_TRADDR) 108 len += scnprintf(buf, size, "traddr=%s", ctrl->opts->traddr); 109 if (ctrl->opts->mask & NVMF_OPT_TRSVCID) 110 len += scnprintf(buf + len, size - len, "%strsvcid=%s", 111 (len) ? "," : "", ctrl->opts->trsvcid); 112 if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR) 113 len += scnprintf(buf + len, size - len, "%shost_traddr=%s", 114 (len) ? "," : "", ctrl->opts->host_traddr); 115 len += scnprintf(buf + len, size - len, "\n"); 116 117 return len; 118 } 119 EXPORT_SYMBOL_GPL(nvmf_get_address); 120 121 /** 122 * nvmf_reg_read32() - NVMe Fabrics "Property Get" API function. 123 * @ctrl: Host NVMe controller instance maintaining the admin 124 * queue used to submit the property read command to 125 * the allocated NVMe controller resource on the target system. 126 * @off: Starting offset value of the targeted property 127 * register (see the fabrics section of the NVMe standard). 128 * @val: OUTPUT parameter that will contain the value of 129 * the property after a successful read. 130 * 131 * Used by the host system to retrieve a 32-bit capsule property value 132 * from an NVMe controller on the target system. 133 * 134 * ("Capsule property" is an "PCIe register concept" applied to the 135 * NVMe fabrics space.) 136 * 137 * Return: 138 * 0: successful read 139 * > 0: NVMe error status code 140 * < 0: Linux errno error code 141 */ 142 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val) 143 { 144 struct nvme_command cmd; 145 union nvme_result res; 146 int ret; 147 148 memset(&cmd, 0, sizeof(cmd)); 149 cmd.prop_get.opcode = nvme_fabrics_command; 150 cmd.prop_get.fctype = nvme_fabrics_type_property_get; 151 cmd.prop_get.offset = cpu_to_le32(off); 152 153 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0, 154 NVME_QID_ANY, 0, 0, false); 155 156 if (ret >= 0) 157 *val = le64_to_cpu(res.u64); 158 if (unlikely(ret != 0)) 159 dev_err(ctrl->device, 160 "Property Get error: %d, offset %#x\n", 161 ret > 0 ? ret & ~NVME_SC_DNR : ret, off); 162 163 return ret; 164 } 165 EXPORT_SYMBOL_GPL(nvmf_reg_read32); 166 167 /** 168 * nvmf_reg_read64() - NVMe Fabrics "Property Get" API function. 169 * @ctrl: Host NVMe controller instance maintaining the admin 170 * queue used to submit the property read command to 171 * the allocated controller resource on the target system. 172 * @off: Starting offset value of the targeted property 173 * register (see the fabrics section of the NVMe standard). 174 * @val: OUTPUT parameter that will contain the value of 175 * the property after a successful read. 176 * 177 * Used by the host system to retrieve a 64-bit capsule property value 178 * from an NVMe controller on the target system. 179 * 180 * ("Capsule property" is an "PCIe register concept" applied to the 181 * NVMe fabrics space.) 182 * 183 * Return: 184 * 0: successful read 185 * > 0: NVMe error status code 186 * < 0: Linux errno error code 187 */ 188 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val) 189 { 190 struct nvme_command cmd; 191 union nvme_result res; 192 int ret; 193 194 memset(&cmd, 0, sizeof(cmd)); 195 cmd.prop_get.opcode = nvme_fabrics_command; 196 cmd.prop_get.fctype = nvme_fabrics_type_property_get; 197 cmd.prop_get.attrib = 1; 198 cmd.prop_get.offset = cpu_to_le32(off); 199 200 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0, 201 NVME_QID_ANY, 0, 0, false); 202 203 if (ret >= 0) 204 *val = le64_to_cpu(res.u64); 205 if (unlikely(ret != 0)) 206 dev_err(ctrl->device, 207 "Property Get error: %d, offset %#x\n", 208 ret > 0 ? ret & ~NVME_SC_DNR : ret, off); 209 return ret; 210 } 211 EXPORT_SYMBOL_GPL(nvmf_reg_read64); 212 213 /** 214 * nvmf_reg_write32() - NVMe Fabrics "Property Write" API function. 215 * @ctrl: Host NVMe controller instance maintaining the admin 216 * queue used to submit the property read command to 217 * the allocated NVMe controller resource on the target system. 218 * @off: Starting offset value of the targeted property 219 * register (see the fabrics section of the NVMe standard). 220 * @val: Input parameter that contains the value to be 221 * written to the property. 222 * 223 * Used by the NVMe host system to write a 32-bit capsule property value 224 * to an NVMe controller on the target system. 225 * 226 * ("Capsule property" is an "PCIe register concept" applied to the 227 * NVMe fabrics space.) 228 * 229 * Return: 230 * 0: successful write 231 * > 0: NVMe error status code 232 * < 0: Linux errno error code 233 */ 234 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val) 235 { 236 struct nvme_command cmd; 237 int ret; 238 239 memset(&cmd, 0, sizeof(cmd)); 240 cmd.prop_set.opcode = nvme_fabrics_command; 241 cmd.prop_set.fctype = nvme_fabrics_type_property_set; 242 cmd.prop_set.attrib = 0; 243 cmd.prop_set.offset = cpu_to_le32(off); 244 cmd.prop_set.value = cpu_to_le64(val); 245 246 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, NULL, NULL, 0, 0, 247 NVME_QID_ANY, 0, 0, false); 248 if (unlikely(ret)) 249 dev_err(ctrl->device, 250 "Property Set error: %d, offset %#x\n", 251 ret > 0 ? ret & ~NVME_SC_DNR : ret, off); 252 return ret; 253 } 254 EXPORT_SYMBOL_GPL(nvmf_reg_write32); 255 256 /** 257 * nvmf_log_connect_error() - Error-parsing-diagnostic print 258 * out function for connect() errors. 259 * 260 * @ctrl: the specific /dev/nvmeX device that had the error. 261 * 262 * @errval: Error code to be decoded in a more human-friendly 263 * printout. 264 * 265 * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM. 266 * 267 * @cmd: This is the SQE portion of a submission capsule. 268 * 269 * @data: This is the "Data" portion of a submission capsule. 270 */ 271 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl, 272 int errval, int offset, struct nvme_command *cmd, 273 struct nvmf_connect_data *data) 274 { 275 int err_sctype = errval & (~NVME_SC_DNR); 276 277 switch (err_sctype) { 278 279 case (NVME_SC_CONNECT_INVALID_PARAM): 280 if (offset >> 16) { 281 char *inv_data = "Connect Invalid Data Parameter"; 282 283 switch (offset & 0xffff) { 284 case (offsetof(struct nvmf_connect_data, cntlid)): 285 dev_err(ctrl->device, 286 "%s, cntlid: %d\n", 287 inv_data, data->cntlid); 288 break; 289 case (offsetof(struct nvmf_connect_data, hostnqn)): 290 dev_err(ctrl->device, 291 "%s, hostnqn \"%s\"\n", 292 inv_data, data->hostnqn); 293 break; 294 case (offsetof(struct nvmf_connect_data, subsysnqn)): 295 dev_err(ctrl->device, 296 "%s, subsysnqn \"%s\"\n", 297 inv_data, data->subsysnqn); 298 break; 299 default: 300 dev_err(ctrl->device, 301 "%s, starting byte offset: %d\n", 302 inv_data, offset & 0xffff); 303 break; 304 } 305 } else { 306 char *inv_sqe = "Connect Invalid SQE Parameter"; 307 308 switch (offset) { 309 case (offsetof(struct nvmf_connect_command, qid)): 310 dev_err(ctrl->device, 311 "%s, qid %d\n", 312 inv_sqe, cmd->connect.qid); 313 break; 314 default: 315 dev_err(ctrl->device, 316 "%s, starting byte offset: %d\n", 317 inv_sqe, offset); 318 } 319 } 320 break; 321 322 case NVME_SC_CONNECT_INVALID_HOST: 323 dev_err(ctrl->device, 324 "Connect for subsystem %s is not allowed, hostnqn: %s\n", 325 data->subsysnqn, data->hostnqn); 326 break; 327 328 case NVME_SC_CONNECT_CTRL_BUSY: 329 dev_err(ctrl->device, 330 "Connect command failed: controller is busy or not available\n"); 331 break; 332 333 case NVME_SC_CONNECT_FORMAT: 334 dev_err(ctrl->device, 335 "Connect incompatible format: %d", 336 cmd->connect.recfmt); 337 break; 338 339 case NVME_SC_HOST_PATH_ERROR: 340 dev_err(ctrl->device, 341 "Connect command failed: host path error\n"); 342 break; 343 344 default: 345 dev_err(ctrl->device, 346 "Connect command failed, error wo/DNR bit: %d\n", 347 err_sctype); 348 break; 349 } /* switch (err_sctype) */ 350 } 351 352 /** 353 * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect" 354 * API function. 355 * @ctrl: Host nvme controller instance used to request 356 * a new NVMe controller allocation on the target 357 * system and establish an NVMe Admin connection to 358 * that controller. 359 * 360 * This function enables an NVMe host device to request a new allocation of 361 * an NVMe controller resource on a target system as well establish a 362 * fabrics-protocol connection of the NVMe Admin queue between the 363 * host system device and the allocated NVMe controller on the 364 * target system via a NVMe Fabrics "Connect" command. 365 * 366 * Return: 367 * 0: success 368 * > 0: NVMe error status code 369 * < 0: Linux errno error code 370 * 371 */ 372 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl) 373 { 374 struct nvme_command cmd; 375 union nvme_result res; 376 struct nvmf_connect_data *data; 377 int ret; 378 379 memset(&cmd, 0, sizeof(cmd)); 380 cmd.connect.opcode = nvme_fabrics_command; 381 cmd.connect.fctype = nvme_fabrics_type_connect; 382 cmd.connect.qid = 0; 383 cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1); 384 385 /* 386 * Set keep-alive timeout in seconds granularity (ms * 1000) 387 */ 388 cmd.connect.kato = cpu_to_le32(ctrl->kato * 1000); 389 390 if (ctrl->opts->disable_sqflow) 391 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW; 392 393 data = kzalloc(sizeof(*data), GFP_KERNEL); 394 if (!data) 395 return -ENOMEM; 396 397 uuid_copy(&data->hostid, &ctrl->opts->host->id); 398 data->cntlid = cpu_to_le16(0xffff); 399 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE); 400 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE); 401 402 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, 403 data, sizeof(*data), 0, NVME_QID_ANY, 1, 404 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT, false); 405 if (ret) { 406 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32), 407 &cmd, data); 408 goto out_free_data; 409 } 410 411 ctrl->cntlid = le16_to_cpu(res.u16); 412 413 out_free_data: 414 kfree(data); 415 return ret; 416 } 417 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue); 418 419 /** 420 * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect" 421 * API function. 422 * @ctrl: Host nvme controller instance used to establish an 423 * NVMe I/O queue connection to the already allocated NVMe 424 * controller on the target system. 425 * @qid: NVMe I/O queue number for the new I/O connection between 426 * host and target (note qid == 0 is illegal as this is 427 * the Admin queue, per NVMe standard). 428 * @poll: Whether or not to poll for the completion of the connect cmd. 429 * 430 * This function issues a fabrics-protocol connection 431 * of a NVMe I/O queue (via NVMe Fabrics "Connect" command) 432 * between the host system device and the allocated NVMe controller 433 * on the target system. 434 * 435 * Return: 436 * 0: success 437 * > 0: NVMe error status code 438 * < 0: Linux errno error code 439 */ 440 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid, bool poll) 441 { 442 struct nvme_command cmd; 443 struct nvmf_connect_data *data; 444 union nvme_result res; 445 int ret; 446 447 memset(&cmd, 0, sizeof(cmd)); 448 cmd.connect.opcode = nvme_fabrics_command; 449 cmd.connect.fctype = nvme_fabrics_type_connect; 450 cmd.connect.qid = cpu_to_le16(qid); 451 cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize); 452 453 if (ctrl->opts->disable_sqflow) 454 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW; 455 456 data = kzalloc(sizeof(*data), GFP_KERNEL); 457 if (!data) 458 return -ENOMEM; 459 460 uuid_copy(&data->hostid, &ctrl->opts->host->id); 461 data->cntlid = cpu_to_le16(ctrl->cntlid); 462 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE); 463 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE); 464 465 ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res, 466 data, sizeof(*data), 0, qid, 1, 467 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT, poll); 468 if (ret) { 469 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32), 470 &cmd, data); 471 } 472 kfree(data); 473 return ret; 474 } 475 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue); 476 477 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl) 478 { 479 if (ctrl->opts->max_reconnects == -1 || 480 ctrl->nr_reconnects < ctrl->opts->max_reconnects) 481 return true; 482 483 return false; 484 } 485 EXPORT_SYMBOL_GPL(nvmf_should_reconnect); 486 487 /** 488 * nvmf_register_transport() - NVMe Fabrics Library registration function. 489 * @ops: Transport ops instance to be registered to the 490 * common fabrics library. 491 * 492 * API function that registers the type of specific transport fabric 493 * being implemented to the common NVMe fabrics library. Part of 494 * the overall init sequence of starting up a fabrics driver. 495 */ 496 int nvmf_register_transport(struct nvmf_transport_ops *ops) 497 { 498 if (!ops->create_ctrl) 499 return -EINVAL; 500 501 down_write(&nvmf_transports_rwsem); 502 list_add_tail(&ops->entry, &nvmf_transports); 503 up_write(&nvmf_transports_rwsem); 504 505 return 0; 506 } 507 EXPORT_SYMBOL_GPL(nvmf_register_transport); 508 509 /** 510 * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function. 511 * @ops: Transport ops instance to be unregistered from the 512 * common fabrics library. 513 * 514 * Fabrics API function that unregisters the type of specific transport 515 * fabric being implemented from the common NVMe fabrics library. 516 * Part of the overall exit sequence of unloading the implemented driver. 517 */ 518 void nvmf_unregister_transport(struct nvmf_transport_ops *ops) 519 { 520 down_write(&nvmf_transports_rwsem); 521 list_del(&ops->entry); 522 up_write(&nvmf_transports_rwsem); 523 } 524 EXPORT_SYMBOL_GPL(nvmf_unregister_transport); 525 526 static struct nvmf_transport_ops *nvmf_lookup_transport( 527 struct nvmf_ctrl_options *opts) 528 { 529 struct nvmf_transport_ops *ops; 530 531 lockdep_assert_held(&nvmf_transports_rwsem); 532 533 list_for_each_entry(ops, &nvmf_transports, entry) { 534 if (strcmp(ops->name, opts->transport) == 0) 535 return ops; 536 } 537 538 return NULL; 539 } 540 541 static const match_table_t opt_tokens = { 542 { NVMF_OPT_TRANSPORT, "transport=%s" }, 543 { NVMF_OPT_TRADDR, "traddr=%s" }, 544 { NVMF_OPT_TRSVCID, "trsvcid=%s" }, 545 { NVMF_OPT_NQN, "nqn=%s" }, 546 { NVMF_OPT_QUEUE_SIZE, "queue_size=%d" }, 547 { NVMF_OPT_NR_IO_QUEUES, "nr_io_queues=%d" }, 548 { NVMF_OPT_RECONNECT_DELAY, "reconnect_delay=%d" }, 549 { NVMF_OPT_CTRL_LOSS_TMO, "ctrl_loss_tmo=%d" }, 550 { NVMF_OPT_KATO, "keep_alive_tmo=%d" }, 551 { NVMF_OPT_HOSTNQN, "hostnqn=%s" }, 552 { NVMF_OPT_HOST_TRADDR, "host_traddr=%s" }, 553 { NVMF_OPT_HOST_ID, "hostid=%s" }, 554 { NVMF_OPT_DUP_CONNECT, "duplicate_connect" }, 555 { NVMF_OPT_DISABLE_SQFLOW, "disable_sqflow" }, 556 { NVMF_OPT_HDR_DIGEST, "hdr_digest" }, 557 { NVMF_OPT_DATA_DIGEST, "data_digest" }, 558 { NVMF_OPT_NR_WRITE_QUEUES, "nr_write_queues=%d" }, 559 { NVMF_OPT_NR_POLL_QUEUES, "nr_poll_queues=%d" }, 560 { NVMF_OPT_TOS, "tos=%d" }, 561 { NVMF_OPT_FAIL_FAST_TMO, "fast_io_fail_tmo=%d" }, 562 { NVMF_OPT_ERR, NULL } 563 }; 564 565 static int nvmf_parse_options(struct nvmf_ctrl_options *opts, 566 const char *buf) 567 { 568 substring_t args[MAX_OPT_ARGS]; 569 char *options, *o, *p; 570 int token, ret = 0; 571 size_t nqnlen = 0; 572 int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO; 573 uuid_t hostid; 574 575 /* Set defaults */ 576 opts->queue_size = NVMF_DEF_QUEUE_SIZE; 577 opts->nr_io_queues = num_online_cpus(); 578 opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY; 579 opts->kato = 0; 580 opts->duplicate_connect = false; 581 opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO; 582 opts->hdr_digest = false; 583 opts->data_digest = false; 584 opts->tos = -1; /* < 0 == use transport default */ 585 586 options = o = kstrdup(buf, GFP_KERNEL); 587 if (!options) 588 return -ENOMEM; 589 590 uuid_gen(&hostid); 591 592 while ((p = strsep(&o, ",\n")) != NULL) { 593 if (!*p) 594 continue; 595 596 token = match_token(p, opt_tokens, args); 597 opts->mask |= token; 598 switch (token) { 599 case NVMF_OPT_TRANSPORT: 600 p = match_strdup(args); 601 if (!p) { 602 ret = -ENOMEM; 603 goto out; 604 } 605 kfree(opts->transport); 606 opts->transport = p; 607 break; 608 case NVMF_OPT_NQN: 609 p = match_strdup(args); 610 if (!p) { 611 ret = -ENOMEM; 612 goto out; 613 } 614 kfree(opts->subsysnqn); 615 opts->subsysnqn = p; 616 nqnlen = strlen(opts->subsysnqn); 617 if (nqnlen >= NVMF_NQN_SIZE) { 618 pr_err("%s needs to be < %d bytes\n", 619 opts->subsysnqn, NVMF_NQN_SIZE); 620 ret = -EINVAL; 621 goto out; 622 } 623 opts->discovery_nqn = 624 !(strcmp(opts->subsysnqn, 625 NVME_DISC_SUBSYS_NAME)); 626 break; 627 case NVMF_OPT_TRADDR: 628 p = match_strdup(args); 629 if (!p) { 630 ret = -ENOMEM; 631 goto out; 632 } 633 kfree(opts->traddr); 634 opts->traddr = p; 635 break; 636 case NVMF_OPT_TRSVCID: 637 p = match_strdup(args); 638 if (!p) { 639 ret = -ENOMEM; 640 goto out; 641 } 642 kfree(opts->trsvcid); 643 opts->trsvcid = p; 644 break; 645 case NVMF_OPT_QUEUE_SIZE: 646 if (match_int(args, &token)) { 647 ret = -EINVAL; 648 goto out; 649 } 650 if (token < NVMF_MIN_QUEUE_SIZE || 651 token > NVMF_MAX_QUEUE_SIZE) { 652 pr_err("Invalid queue_size %d\n", token); 653 ret = -EINVAL; 654 goto out; 655 } 656 opts->queue_size = token; 657 break; 658 case NVMF_OPT_NR_IO_QUEUES: 659 if (match_int(args, &token)) { 660 ret = -EINVAL; 661 goto out; 662 } 663 if (token <= 0) { 664 pr_err("Invalid number of IOQs %d\n", token); 665 ret = -EINVAL; 666 goto out; 667 } 668 if (opts->discovery_nqn) { 669 pr_debug("Ignoring nr_io_queues value for discovery controller\n"); 670 break; 671 } 672 673 opts->nr_io_queues = min_t(unsigned int, 674 num_online_cpus(), token); 675 break; 676 case NVMF_OPT_KATO: 677 if (match_int(args, &token)) { 678 ret = -EINVAL; 679 goto out; 680 } 681 682 if (token < 0) { 683 pr_err("Invalid keep_alive_tmo %d\n", token); 684 ret = -EINVAL; 685 goto out; 686 } else if (token == 0 && !opts->discovery_nqn) { 687 /* Allowed for debug */ 688 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n"); 689 } 690 opts->kato = token; 691 break; 692 case NVMF_OPT_CTRL_LOSS_TMO: 693 if (match_int(args, &token)) { 694 ret = -EINVAL; 695 goto out; 696 } 697 698 if (token < 0) 699 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n"); 700 ctrl_loss_tmo = token; 701 break; 702 case NVMF_OPT_FAIL_FAST_TMO: 703 if (match_int(args, &token)) { 704 ret = -EINVAL; 705 goto out; 706 } 707 708 if (token >= 0) 709 pr_warn("I/O fail on reconnect controller after %d sec\n", 710 token); 711 opts->fast_io_fail_tmo = token; 712 break; 713 case NVMF_OPT_HOSTNQN: 714 if (opts->host) { 715 pr_err("hostnqn already user-assigned: %s\n", 716 opts->host->nqn); 717 ret = -EADDRINUSE; 718 goto out; 719 } 720 p = match_strdup(args); 721 if (!p) { 722 ret = -ENOMEM; 723 goto out; 724 } 725 nqnlen = strlen(p); 726 if (nqnlen >= NVMF_NQN_SIZE) { 727 pr_err("%s needs to be < %d bytes\n", 728 p, NVMF_NQN_SIZE); 729 kfree(p); 730 ret = -EINVAL; 731 goto out; 732 } 733 nvmf_host_put(opts->host); 734 opts->host = nvmf_host_add(p); 735 kfree(p); 736 if (!opts->host) { 737 ret = -ENOMEM; 738 goto out; 739 } 740 break; 741 case NVMF_OPT_RECONNECT_DELAY: 742 if (match_int(args, &token)) { 743 ret = -EINVAL; 744 goto out; 745 } 746 if (token <= 0) { 747 pr_err("Invalid reconnect_delay %d\n", token); 748 ret = -EINVAL; 749 goto out; 750 } 751 opts->reconnect_delay = token; 752 break; 753 case NVMF_OPT_HOST_TRADDR: 754 p = match_strdup(args); 755 if (!p) { 756 ret = -ENOMEM; 757 goto out; 758 } 759 kfree(opts->host_traddr); 760 opts->host_traddr = p; 761 break; 762 case NVMF_OPT_HOST_ID: 763 p = match_strdup(args); 764 if (!p) { 765 ret = -ENOMEM; 766 goto out; 767 } 768 ret = uuid_parse(p, &hostid); 769 if (ret) { 770 pr_err("Invalid hostid %s\n", p); 771 ret = -EINVAL; 772 kfree(p); 773 goto out; 774 } 775 kfree(p); 776 break; 777 case NVMF_OPT_DUP_CONNECT: 778 opts->duplicate_connect = true; 779 break; 780 case NVMF_OPT_DISABLE_SQFLOW: 781 opts->disable_sqflow = true; 782 break; 783 case NVMF_OPT_HDR_DIGEST: 784 opts->hdr_digest = true; 785 break; 786 case NVMF_OPT_DATA_DIGEST: 787 opts->data_digest = true; 788 break; 789 case NVMF_OPT_NR_WRITE_QUEUES: 790 if (match_int(args, &token)) { 791 ret = -EINVAL; 792 goto out; 793 } 794 if (token <= 0) { 795 pr_err("Invalid nr_write_queues %d\n", token); 796 ret = -EINVAL; 797 goto out; 798 } 799 opts->nr_write_queues = token; 800 break; 801 case NVMF_OPT_NR_POLL_QUEUES: 802 if (match_int(args, &token)) { 803 ret = -EINVAL; 804 goto out; 805 } 806 if (token <= 0) { 807 pr_err("Invalid nr_poll_queues %d\n", token); 808 ret = -EINVAL; 809 goto out; 810 } 811 opts->nr_poll_queues = token; 812 break; 813 case NVMF_OPT_TOS: 814 if (match_int(args, &token)) { 815 ret = -EINVAL; 816 goto out; 817 } 818 if (token < 0) { 819 pr_err("Invalid type of service %d\n", token); 820 ret = -EINVAL; 821 goto out; 822 } 823 if (token > 255) { 824 pr_warn("Clamping type of service to 255\n"); 825 token = 255; 826 } 827 opts->tos = token; 828 break; 829 default: 830 pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n", 831 p); 832 ret = -EINVAL; 833 goto out; 834 } 835 } 836 837 if (opts->discovery_nqn) { 838 opts->nr_io_queues = 0; 839 opts->nr_write_queues = 0; 840 opts->nr_poll_queues = 0; 841 opts->duplicate_connect = true; 842 } else { 843 if (!opts->kato) 844 opts->kato = NVME_DEFAULT_KATO; 845 } 846 if (ctrl_loss_tmo < 0) { 847 opts->max_reconnects = -1; 848 } else { 849 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo, 850 opts->reconnect_delay); 851 if (ctrl_loss_tmo < opts->fast_io_fail_tmo) 852 pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n", 853 opts->fast_io_fail_tmo, ctrl_loss_tmo); 854 } 855 856 if (!opts->host) { 857 kref_get(&nvmf_default_host->ref); 858 opts->host = nvmf_default_host; 859 } 860 861 uuid_copy(&opts->host->id, &hostid); 862 863 out: 864 kfree(options); 865 return ret; 866 } 867 868 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts, 869 unsigned int required_opts) 870 { 871 if ((opts->mask & required_opts) != required_opts) { 872 int i; 873 874 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) { 875 if ((opt_tokens[i].token & required_opts) && 876 !(opt_tokens[i].token & opts->mask)) { 877 pr_warn("missing parameter '%s'\n", 878 opt_tokens[i].pattern); 879 } 880 } 881 882 return -EINVAL; 883 } 884 885 return 0; 886 } 887 888 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl, 889 struct nvmf_ctrl_options *opts) 890 { 891 if (!nvmf_ctlr_matches_baseopts(ctrl, opts) || 892 strcmp(opts->traddr, ctrl->opts->traddr) || 893 strcmp(opts->trsvcid, ctrl->opts->trsvcid)) 894 return false; 895 896 /* 897 * Checking the local address is rough. In most cases, none is specified 898 * and the host port is selected by the stack. 899 * 900 * Assume no match if: 901 * - local address is specified and address is not the same 902 * - local address is not specified but remote is, or vice versa 903 * (admin using specific host_traddr when it matters). 904 */ 905 if ((opts->mask & NVMF_OPT_HOST_TRADDR) && 906 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) { 907 if (strcmp(opts->host_traddr, ctrl->opts->host_traddr)) 908 return false; 909 } else if ((opts->mask & NVMF_OPT_HOST_TRADDR) || 910 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) { 911 return false; 912 } 913 914 return true; 915 } 916 EXPORT_SYMBOL_GPL(nvmf_ip_options_match); 917 918 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts, 919 unsigned int allowed_opts) 920 { 921 if (opts->mask & ~allowed_opts) { 922 int i; 923 924 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) { 925 if ((opt_tokens[i].token & opts->mask) && 926 (opt_tokens[i].token & ~allowed_opts)) { 927 pr_warn("invalid parameter '%s'\n", 928 opt_tokens[i].pattern); 929 } 930 } 931 932 return -EINVAL; 933 } 934 935 return 0; 936 } 937 938 void nvmf_free_options(struct nvmf_ctrl_options *opts) 939 { 940 nvmf_host_put(opts->host); 941 kfree(opts->transport); 942 kfree(opts->traddr); 943 kfree(opts->trsvcid); 944 kfree(opts->subsysnqn); 945 kfree(opts->host_traddr); 946 kfree(opts); 947 } 948 EXPORT_SYMBOL_GPL(nvmf_free_options); 949 950 #define NVMF_REQUIRED_OPTS (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN) 951 #define NVMF_ALLOWED_OPTS (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \ 952 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \ 953 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\ 954 NVMF_OPT_DISABLE_SQFLOW |\ 955 NVMF_OPT_FAIL_FAST_TMO) 956 957 static struct nvme_ctrl * 958 nvmf_create_ctrl(struct device *dev, const char *buf) 959 { 960 struct nvmf_ctrl_options *opts; 961 struct nvmf_transport_ops *ops; 962 struct nvme_ctrl *ctrl; 963 int ret; 964 965 opts = kzalloc(sizeof(*opts), GFP_KERNEL); 966 if (!opts) 967 return ERR_PTR(-ENOMEM); 968 969 ret = nvmf_parse_options(opts, buf); 970 if (ret) 971 goto out_free_opts; 972 973 974 request_module("nvme-%s", opts->transport); 975 976 /* 977 * Check the generic options first as we need a valid transport for 978 * the lookup below. Then clear the generic flags so that transport 979 * drivers don't have to care about them. 980 */ 981 ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS); 982 if (ret) 983 goto out_free_opts; 984 opts->mask &= ~NVMF_REQUIRED_OPTS; 985 986 down_read(&nvmf_transports_rwsem); 987 ops = nvmf_lookup_transport(opts); 988 if (!ops) { 989 pr_info("no handler found for transport %s.\n", 990 opts->transport); 991 ret = -EINVAL; 992 goto out_unlock; 993 } 994 995 if (!try_module_get(ops->module)) { 996 ret = -EBUSY; 997 goto out_unlock; 998 } 999 up_read(&nvmf_transports_rwsem); 1000 1001 ret = nvmf_check_required_opts(opts, ops->required_opts); 1002 if (ret) 1003 goto out_module_put; 1004 ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS | 1005 ops->allowed_opts | ops->required_opts); 1006 if (ret) 1007 goto out_module_put; 1008 1009 ctrl = ops->create_ctrl(dev, opts); 1010 if (IS_ERR(ctrl)) { 1011 ret = PTR_ERR(ctrl); 1012 goto out_module_put; 1013 } 1014 1015 module_put(ops->module); 1016 return ctrl; 1017 1018 out_module_put: 1019 module_put(ops->module); 1020 goto out_free_opts; 1021 out_unlock: 1022 up_read(&nvmf_transports_rwsem); 1023 out_free_opts: 1024 nvmf_free_options(opts); 1025 return ERR_PTR(ret); 1026 } 1027 1028 static struct class *nvmf_class; 1029 static struct device *nvmf_device; 1030 static DEFINE_MUTEX(nvmf_dev_mutex); 1031 1032 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf, 1033 size_t count, loff_t *pos) 1034 { 1035 struct seq_file *seq_file = file->private_data; 1036 struct nvme_ctrl *ctrl; 1037 const char *buf; 1038 int ret = 0; 1039 1040 if (count > PAGE_SIZE) 1041 return -ENOMEM; 1042 1043 buf = memdup_user_nul(ubuf, count); 1044 if (IS_ERR(buf)) 1045 return PTR_ERR(buf); 1046 1047 mutex_lock(&nvmf_dev_mutex); 1048 if (seq_file->private) { 1049 ret = -EINVAL; 1050 goto out_unlock; 1051 } 1052 1053 ctrl = nvmf_create_ctrl(nvmf_device, buf); 1054 if (IS_ERR(ctrl)) { 1055 ret = PTR_ERR(ctrl); 1056 goto out_unlock; 1057 } 1058 1059 seq_file->private = ctrl; 1060 1061 out_unlock: 1062 mutex_unlock(&nvmf_dev_mutex); 1063 kfree(buf); 1064 return ret ? ret : count; 1065 } 1066 1067 static int nvmf_dev_show(struct seq_file *seq_file, void *private) 1068 { 1069 struct nvme_ctrl *ctrl; 1070 int ret = 0; 1071 1072 mutex_lock(&nvmf_dev_mutex); 1073 ctrl = seq_file->private; 1074 if (!ctrl) { 1075 ret = -EINVAL; 1076 goto out_unlock; 1077 } 1078 1079 seq_printf(seq_file, "instance=%d,cntlid=%d\n", 1080 ctrl->instance, ctrl->cntlid); 1081 1082 out_unlock: 1083 mutex_unlock(&nvmf_dev_mutex); 1084 return ret; 1085 } 1086 1087 static int nvmf_dev_open(struct inode *inode, struct file *file) 1088 { 1089 /* 1090 * The miscdevice code initializes file->private_data, but doesn't 1091 * make use of it later. 1092 */ 1093 file->private_data = NULL; 1094 return single_open(file, nvmf_dev_show, NULL); 1095 } 1096 1097 static int nvmf_dev_release(struct inode *inode, struct file *file) 1098 { 1099 struct seq_file *seq_file = file->private_data; 1100 struct nvme_ctrl *ctrl = seq_file->private; 1101 1102 if (ctrl) 1103 nvme_put_ctrl(ctrl); 1104 return single_release(inode, file); 1105 } 1106 1107 static const struct file_operations nvmf_dev_fops = { 1108 .owner = THIS_MODULE, 1109 .write = nvmf_dev_write, 1110 .read = seq_read, 1111 .open = nvmf_dev_open, 1112 .release = nvmf_dev_release, 1113 }; 1114 1115 static struct miscdevice nvmf_misc = { 1116 .minor = MISC_DYNAMIC_MINOR, 1117 .name = "nvme-fabrics", 1118 .fops = &nvmf_dev_fops, 1119 }; 1120 1121 static int __init nvmf_init(void) 1122 { 1123 int ret; 1124 1125 nvmf_default_host = nvmf_host_default(); 1126 if (!nvmf_default_host) 1127 return -ENOMEM; 1128 1129 nvmf_class = class_create(THIS_MODULE, "nvme-fabrics"); 1130 if (IS_ERR(nvmf_class)) { 1131 pr_err("couldn't register class nvme-fabrics\n"); 1132 ret = PTR_ERR(nvmf_class); 1133 goto out_free_host; 1134 } 1135 1136 nvmf_device = 1137 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl"); 1138 if (IS_ERR(nvmf_device)) { 1139 pr_err("couldn't create nvme-fabris device!\n"); 1140 ret = PTR_ERR(nvmf_device); 1141 goto out_destroy_class; 1142 } 1143 1144 ret = misc_register(&nvmf_misc); 1145 if (ret) { 1146 pr_err("couldn't register misc device: %d\n", ret); 1147 goto out_destroy_device; 1148 } 1149 1150 return 0; 1151 1152 out_destroy_device: 1153 device_destroy(nvmf_class, MKDEV(0, 0)); 1154 out_destroy_class: 1155 class_destroy(nvmf_class); 1156 out_free_host: 1157 nvmf_host_put(nvmf_default_host); 1158 return ret; 1159 } 1160 1161 static void __exit nvmf_exit(void) 1162 { 1163 misc_deregister(&nvmf_misc); 1164 device_destroy(nvmf_class, MKDEV(0, 0)); 1165 class_destroy(nvmf_class); 1166 nvmf_host_put(nvmf_default_host); 1167 1168 BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64); 1169 BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64); 1170 BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64); 1171 BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64); 1172 BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024); 1173 } 1174 1175 MODULE_LICENSE("GPL v2"); 1176 1177 module_init(nvmf_init); 1178 module_exit(nvmf_exit); 1179